The Use of AI in Creating 3D Models for 3D Printing

By Jack Martinich, Sales & Applications, Formero

One of the most common conversations I have with customers starts with a great idea and ends with a pause. They know what they want to build. They just don’t have a 3D file to start from. For years, the answer was straightforward: find a designer, commission a model, and then come back to us. That process could take weeks and cost significant money before a single part had even been printed.

AI is starting to change that. They’re not perfect and come with pitfalls, but the tools have improved enough that they are worth taking seriously. Here is an honest look at where things stand.

What these tools actually do

A new category of AI-powered design tools has emerged over the past few years, and the pace of development is striking. These platforms let you describe what you want in plain language or upload a reference image, and generate a 3D model in seconds. No CAD software. No modelling experience required.

Some of the platforms that are being discussed and may be worth trying include:

• Adam (adam.new)
• Meshy (meshy.ai)
• Tripo AI (tripo3d.ai)
• Printpal (printpal.io)

These are not the only options, and the space is evolving quickly. But they represent the current accessible state of the art for someone who wants to go from idea to printable file with no CAD background.

Where AI genuinely adds value

In my experience, AI tools offer the most benefit at the two extremes of the design complexity.

At the high end, there are forms that are genuinely difficult or time-consuming for a human designer to model. Complex organic geometry, intricate surface textures, and topology-optimised structures that would take days in conventional CAD software can be approximated by AI tools in minutes. Additive manufacturing is particularly well-suited to these kinds of forms. Unlike CNC machining or injection moulding, 3D printing does not penalise complexity in the same way. A part that would be prohibitively expensive to machine can often be printed without difficulty, and if AI can help generate that geometry in the first place, the combination is powerful.

At the other end, there are parts that are simply too straightforward to justify the cost of professional design. A basic bracket, a simple enclosure, a low-tolerance custom fitting. If the part is not safety-critical, does not require tight dimensional accuracy, and can tolerate some geometric imperfection, an AI-generated model might be entirely good enough. Getting to a usable file in minutes instead of days has real value, even if the result is not perfect.

The messy middle, where precision matters and the design is moderately complex, is where AI may still struggle. That is still the domain of skilled designers working in proper CAD environments. I do not think that will change soon.

What you need to understand before you rely on an AI-generated file

This is where I want to be direct, because the limitations are real and they matter for 3D printing specifically.

Most AI tools output mesh files, typically in STL or OBJ format. Mesh files are what 3D printers work with, but they are also inherently prone to errors. A model that looks correct on screen can contain thin walls that cannot physically exist, zero-thickness surfaces, gaps or holes in the geometry, and surface artefacts that will affect the final part. If you send us a file with any of these issues, we will either flag it before printing or you will receive a part that does not match your expectations.

The underlying issue is that AI tools generate geometry visually. They are optimising for how a 3D model looks, suitable for a virtual asset or 3D rendering, but not necessarily for whether it can exist in the real world. The better tools are getting smarter about this, and platforms like Adam and Printpal explicitly design toward printability, but the problem has not gone away.

There is also a broader constraint worth understanding. Producing a design using AI means handing over a degree of direct control. When a skilled designer builds a parametric CAD model, every dimension is intentional and adjustable. An AI-generated mesh is much harder to edit with precision after the fact. For projects where tight tolerances are critical, where parts need to interface accurately with other components, or where the design will eventually move to CNC machining or injection moulding, a mesh-based AI output is likely to create more problems than it solves. Those processes rely on BREP manufacturing file formats and 2D drawings, not mesh geometry, and the conversion between them introduces significant risk.

3D scanning is a related area where AI is already adding meaningful value. Scanning technology can generate enormous amounts of point cloud data, and AI tools are becoming capable of processing and manipulating that data far faster than traditional workflows allow. If you need a part that conforms to a person’s unique anatomy, fits an existing vehicle interior, or matches a legacy component that no longer has a digital model, the combination of 3D scanning and AI processing can be genuinely transformative. Orthotics are a clear example. So are aftermarket automotive parts that need to fit a specific body panel.

Where this is heading

AI is not going to replace skilled industrial designers anytime soon. But I think it is going to become a standard part of the toolkit, in the same way that finite element analysis software or rendering tools did. Not a replacement for expertise, but something that reduces friction and opens up possibilities that were previously too costly or time-consuming to pursue.

Topology optimisation is one area I am watching closely. The idea of applying an AI-driven “skeletonisation” pass to a part after it has been designed, reducing material and mass while maintaining structural performance, is already technically possible. As tools mature, I expect this to become a standard post-processing step for additively manufactured components, particularly in aerospace, automotive, and medical applications where weight and material costs are significant.

Additive manufacturing is also, by its nature, more forgiving of the kinds of organic and unconventional geometry that AI tends to produce. Conventional manufacturing processes impose strict constraints on what geometry is feasible. 3D printing is far more permissive. That compatibility means AI-generated designs and additive manufacturing are natural partners, and I expect that relationship to deepen as both technologies develop.

A practical note for customers without a file

If you have a 3D printing idea but no 3D model, our advice is to try one of the tools listed above. Many offer free tiers and the quality of output has improved enormously in a short period of time. Treat any AI-generated file as a starting point, for example to cost out a project, rather than a finished product. Check it for obvious issues like thin walls and gaps, and when in doubt, send it through. We would rather review a file early than have you disappointed with the result.

We are continuing to evaluate these platforms ourselves and will be sharing more guidance soon, including a practical help guide for customers navigating this process for the first time.

Need a hand getting there?

If you are stuck at the file stage, our team may be able to help. Whether you need help finalising a model, adjusting geometry so it will print successfully, or working out whether your file is ready to quote, get in touch at sales@formero.com.au.

If you are ready to go, shop.formero.com.au is the fastest way to get a free instant quote and get started.

Speak with us today.

3D Printing, Knowledge Hub